For service businesses that miss calls while crews are on jobs, an AI phone agent is an attractive idea. It answers every call, asks the standard questions and books the work. Owners are exploring it: in Homebase's survey, AI-powered customer service use among adopters ranged from 23% in retail to 53% in computer hardware and software.[2] The technology can be good. The risks are in the details, so ask these seven questions before you sign.
1. What happens when it cannot handle the call?
Every voice agent hits calls it cannot handle: an angry customer, a technical question, a bad connection. Ask how it transfers to a person, whether the transfer is warm (the human gets context) or cold, what happens after hours and how long a caller can be stuck in a loop. A caller who cannot reach a human will hang up and tell people.
2. What is it allowed to say and promise?
Get a written list. Can it quote prices? Promise arrival windows? Offer discounts? Confirm warranty coverage? Your safest rule is that it collects information and books standard appointments, and everything else goes to your team. Test its limits by asking leading questions.
3. How does it handle emergencies?
For plumbing, HVAC, electrical, restoration and similar trades, some callers are describing a danger: gas smell, flooding, no heat in winter, sparks. The system needs clear rules to recognize these and transfer immediately, ideally with a text alert to your on-call person. Ask to see the rules and test them.
4. What about recording and disclosure?
Rules on recording calls and on disclosing that the caller is speaking to an automated system vary by state. Ask the vendor what it does by default and what you must configure. Confirm with your attorney, especially if you serve customers in more than one state.
5. How does it connect to your systems?
A voice agent that cannot see your calendar, customer records or dispatch board creates double work: someone has to re-enter every booking. Ask which systems it integrates with natively, which need custom work, and who is responsible when an integration breaks. Check that it can write to the tools you use, and not just email you a summary.
6. How will you review what it did?
You need recordings or transcripts, a way to flag bad calls and a regular review habit. Plan to listen to a sample every day for the first two weeks, then weekly. Ask whether you can see how often the agent transferred, failed to book or got a correction from the caller.
7. What does it cost to leave, and how does cost scale?
Ask how pricing is calculated (per minute, per call, per seat, a base fee plus usage) and what happens in your busiest month. Ask whether you own your call data and can export it, and whether you can keep your phone number if you switch. Do not rely on a list price. Ask for an estimate based on your own call volume.
A test plan before you go live
- Write 20 realistic call scripts from your own call history, including a rushed caller, someone who rambles, a wrong number, a price shopper, an emergency, an upset customer and a caller with an accent or background noise.
- Have people who do not work for you place the calls.
- Score each call: answered correctly, booked correctly, transferred correctly or failed.
- Review the failures with the vendor and fix them before launch.
- Go live on after-hours calls first, where the alternative is voicemail, and expand only when the results hold.
Is it worth it?
Size the problem first. Our article on what a missed call costs includes a calculator. If the value of missed calls is small, you do not need this. If it is large, compare the voice agent against simpler fixes: a text-back, an answering service or a rotating on-call schedule. A voice agent should beat those on cost or booking rate, not just on novelty. Bluevine found that 27% of small business owners were using AI for customer service,[1] so you will find vendors with real customers to talk to. Ask for references from a business your size and in your trade.
Sources
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.